Anthropic's Claude AI achieves breakthrough in protein design for drug discovery

Anthropic has demonstrated that its Claude AI models can successfully perform de novo protein design, a critical early stage in drug discovery. In recent experiments, Claude models including Mythos Preview and Opus 4.8 were tasked with designing protein binders from scratch for 15 different targets. Laboratory testing conducted by independent firms Adaptyv Bio and Twist Bioscience confirmed that the AI successfully designed binders for 14 of the 15 targets.

Anthropic reported hit rates significantly higher than the industry standard of 10% to 15%. Specifically, the Mythos Preview model achieved a hit rate of 35.1% in single-target modes. The AI utilized a variety of open-source specialized software to manage the protein design stack, including PXDesign and SolubleMPNN, rather than relying on a proprietary protein model.

While the results show promise for accelerating biotech research, experts note that the designs currently exhibit relatively low affinity and are still in the early research stages. Additionally, the company has faced recent operational challenges, including a series of service disruptions and elevated error rates across its Claude product line in August 2026.

On the business front, Anthropic is aggressively scaling its infrastructure. The company reportedly holds between 70 and 80 letters of intent for data center capacity as it seeks to support its growing computational needs, following a significant valuation increase.